The machine learning built into these softwares aren't as great as they can be, because you're usually running them on low powered hardware.
I run Immich, and it gets results like this wrong.
Google photos, on the other hand, is insane. I can type in anyone's name, and it'll pull up photos of them when they were a baby, despite those photos being scanned in and have no meta data connection. Also pulls up videos with them inside.
I can search for any text, item, and even break it down to a very specific search such as "blue car with black hood", or, "James and Jack at the beach", and it'll show me photos with both James AND Jack only at the beach.
It's flipping nuts. Hard to compare to some free machine learning half baked into a free open source photo gallery maintained by half a guy, built to run on a raspberry Pi.
Oh yeah of course, even in this case where it got some easy stuff wrong I'm still astounded of what it did get right in other situations; considering as you said this is running on a Pi 4 (8GB)
The machine learning built into these softwares aren't as great as they can be, because you're usually running them on low powered hardware.
I run Immich, and it gets results like this wrong.
Google photos, on the other hand, is insane. I can type in anyone's name, and it'll pull up photos of them when they were a baby, despite those photos being scanned in and have no meta data connection. Also pulls up videos with them inside.
I can search for any text, item, and even break it down to a very specific search such as "blue car with black hood", or, "James and Jack at the beach", and it'll show me photos with both James AND Jack only at the beach.
It's flipping nuts. Hard to compare to some free machine learning half baked into a free open source photo gallery maintained by half a guy, built to run on a raspberry Pi.
Oh yeah of course, even in this case where it got some easy stuff wrong I'm still astounded of what it did get right in other situations; considering as you said this is running on a Pi 4 (8GB)